A dodecylbenzenesulfonic acid sodium salt (CAS 25155–3-0) product mix of homologues and isomers, with ∼80% LAS (cat no. D-2525) obtained from Sigma-Aldrich (Steinheim, Germany), was used. Trihexylamine (THA) and di-n-hyxylether (DHE) (>97%, purum) were obtained from Fluka Chemical AG (Buchs, Switzerland). THA >97% was used for initial studies while purum grade (>99%) was used for quantitative experiments. Acetonitrile, Chromasolv®), SDS (>99%) and humic acids (no. H1,675–2) were obtained from Sigma-Aldrich. Formaldehyde solution (>37%), HCl (37%), Na₂HPO_4, NaH₂PO₄ and NaClO₄ were obtained from Merck (Darmstadt, Germany). NaOH (reagent grade) was obtained from Scharlau Chemie S.A (Barcelona, Spain). NaCl of analytical/pro analysi grade from both Fisher Scientific (Loughborough, UK) and Merck were used. Stock solutions were prepared at 1 g L⁻¹ LAS in methanol. THA stock solutions were prepared at 100 mg L⁻¹ in 0.1 M phosphate buffer at pH 7. SDS was dissolved in 50% water and 50% methanol. All water used in preparation of solutions was purified with a Milli-Q-RO4 system (Millipore, Bedford, MA, USA).

The glassware was burned 10–20 h at 200 °C, cleaned with methanol and kept aside for this project. Between each extraction, sample flasks were cleaned with water and methanol and weekly burned in order to minimize carry-over. Acceptor solutions were NaOH at pH 12 or phosphate buffer adjusted with NaOH to pH 12. Acceptor and calibration solutions were prepared with 200 mg L⁻¹ SDS in order to decrease analyte adsorption. Sample buffers at pH 7.0 were prepared by mixing 10 mM Na₂HPO₄ with 10 mM NaH₂PO₄. A model 211 microprocessor pH meter (Hanna Instruments) was used in the preparation of sample, acceptor and mobile phase buffers. For measuring acceptor pH (in some cases), a Ross 8220BNWP (Thermo Fisher Scientific, Beverly, MA, USA) microelectrode was used.

Q3/2 Accurel PP polypropylene hollow fiber membranes (200 μm wall thickness, 600 μm ID, 66% porosity, 0.2 μm pore size) were obtained from Membrana GmbH (Wuppertal, Germany). Fibers were cut with a scalpel into 3.7 cm long pieces. One end was closed by heating. Fibers were washed with acetone under sonication, dried and kept in a closed Petri dish until use. After this preparation, an HF had an effective length of ∼3.5 cm and could fit ∼10 μL acceptor in the lumen. With a microliter syringe, acceptor buffer was filled into the HF until aqueous solution was seen exiting the pores. The fiber was dipped into DHE for a few minutes to impregnate the pores of the HF wall and form the SLM. To wash surplus solvent from the HF surface, the HF was immersed into water and either shaken up to 30 s or sonicated ∼1 s.

The HF was held in the sample using two slightly different setups. In initial studies, the syringe used to fill the acceptor was carefully replaced with a solid metal rod (soldering tin or stainless steel). The rod could easily be bent so that the HF could be adjusted by the edge of an Erlenmeyer flask or in the neck of a volumetric flask, assuring the HF was below the solution surface and did not interfere with the stir bar magnet. However, this setup had disadvantages, such as increased labor and risk for loss of acceptor, which could occur when the syringe was taken off, the metal rod was inserted or removed or when the syringe was connected to collect the extract (described below). The setup with metal rods was used in studying the effects of counter-ion concentration, extraction time, acceptor composition, sample volume, stirring speed, sodium chloride and application to tap and surface waters, but was eventually replaced. With more microliter syringes at hand, the HF needed not to be removed from the syringe that was used to fill the acceptor. With this setup, i.e. the usual setup for LPME [23,28], the sample was kept in a volumetric flask and the HF was placed a few mm under the solution surface in the neck of the flask. The syringe with the HF was held in place by a laboratory clamp. The syringe setup was used in studying the effects of stirring speed, extraction linearity and matrix effects, such as effects of humic acids and application to surface water.

During extraction, samples were stirred with an IKAMAG RO10 power stirrer (IKA-Werke, Staufen, Germany) with place for 10 samples for simultaneous treatment and maximum speed of 1100 rpm. After extraction, the HF was removed from the sample and quickly blotted with a Kleenex tissue to remove drops on the HF surface. The sealed end of the fiber was cut. The acceptor was withdrawn with a syringe (in applicable cases with the syringe that held the HF during the extraction) and the volume (typically 9–11 μL) was noted. The extract was placed in a vial with a conical insert. For column compatibility, the extract was neutralized with 0.1 M HCl. The volume of HCl was 50% of the extract volume. Capped vials were stored in refrigerator or analyzed directly. To avoid cross contamination, the syringe was washed 3 times in acceptor buffer followed by 3 times in methanol between extracts.

Experiments were evaluated by calculation of extraction efficiencies (E) and enrichment factors (E_e). E is defined as n_A/n_S, where n is the total analyte amount in the sample (subscript S) or in the acceptor (subscript A). E_e, is defined as C_A/C_S. C_S is the nominal spiked sample concentration and C_A the acceptor concentration. Relative matrix effects are described by comparing the enrichment factor in a sample with the studied matrix component (e.g., humic acids) to that of a sample without this matrix component by division, i.e., E_{e-matrix relative} = E_{e-with matrix}/E_{e-without matrix}.

The employed HPLC system was an Agilent 1100 liquid chromatograph (Agilent, Santa Clara, CA, USA). 5 μL was injected with the autosampler. Carry-over was eliminated by washing the needle after each injection with acceptor buffer containing 200 mg L⁻¹ SDS. A C₁₈ column (3 μm, 4.6 mm ID × 150 mm (ACE ®, Aberdeen, Scotland)) was used, giving the possibility to separate both homologues and isomers [19,29,30], even though keeping the isomers unresolved with C₈ columns can increase the homologue signal relative to the baseline [31]. Different mobile phases for LAS separation have been presented, whereof various mixtures of an aqueous phase and acetonitrile have frequently been used [16,17,19,29,30], sometimes with sodium perchlorate [16,19,29] and sometimes in gradient mode [16,17,19]. Separation was performed with acetonitrile and phosphate buffer (5 mM, pH 6). 0.1 M sodium perchlorate was used as modifier to increase resolution [29]. The flow was 0.5 mL min⁻¹ and the column was thermostated to 25 °C. The separation program developed here was as follows: constant 50% for 1 min, gradient to 60% acetonitrile until 10 min, and constant until 28 min. The column was then washed with a 1 min gradient to 70% acetonitrile and kept constant at 70% during at least 2 min, followed by restoration to initial conditions during 1 min and constant at 50% for 1 min. The Agilent 1100 FLD module was used for fluorescence detection. Following the FLD optimization procedure described by Agilent, 230 nm was used for excitation and 310 nm for emission.

Calibration solutions were prepared by dilution of the LAS product with acceptor buffer. The LOD of the HPLC was ∼200 μg L⁻¹. The HPLC was calibrated in two intervals. The low interval was nominally 0.2–30 mg L⁻¹ with the FLD photo-multiplier set to the maximum of 18 (arbitrary units). The high interval was 80–900 mg L⁻¹ with the photo-multiplier set to 13. Isomer peaks were integrated individually, but were summed to obtain the total signal of each homologue, giving linear R²-coefficients of 0.995–0.999. The fractions of homologues in the employed LAS product were estimated to 13% C₁₀, 24% C₁₁, 24% C₁₂ and 19% C₁₃, using the 0.2–30 mg L⁻¹ calibration and assuming equal FLD sensitivity for all isomers as well as exactly 80% of total LAS mass concentration. These fractions were taken into account to estimate LOD for the individual homologues. For calculation of E and E_e, these fractions were not taken into account, since sample, acceptor and calibration solutions were all expected to be equally proportionally lower. Unless otherwise noted, presented LAS concentrations refer to nominal concentrations.

Effluent samples were collected in the wintertime from Källby STP (Lund, Sweden) at the beginning of the third denitrification pond. Recipient river [32] samples were collected ∼1.4 km upstream and ∼1.5 km downstream the plant, respectively. Clean bottles (1–2 L) were dipped upside down into the water and turned toward the direction of water flow. Samples were conserved by adding formaldehyde to a final 1% (v/v) concentration and stored at 4 °C in darkness less than 4 days [12]. In total 54 mL formaldehyde was added per 2 L sample.

In order to maximize enrichment, the ion-pairing amine should be in excess. For extraction of 1 mg L⁻¹ LAS, the tested THA concentrations were 10, 50, 75 and 100 mg L⁻¹ (n = 3). 100 mL samples were extracted during 24 h with 330 rpm stirring and 0.01 M NaOH was used as acceptor. For 50 and 75 mg L⁻¹, precision was poor. For 100 mg L⁻¹, E_e was significantly lower than for 10 mg L⁻¹. Possibly, the highest concentration of THA had a solvating of effect on the SLM during long extractions. 10 mg L⁻¹ THA was considered optimal.

An equilibrium time of 15 h can be observed in Figure 1. In the equilibrium regime, the enrichment is determined by a distribution ratio between the acceptor and the sample, and extraction time is not critical. Unless otherwise noted, 15 h was employed for further experiments. For some sets of experiments, 20 h was used. These extraction times are longer than in typical SPE methods, but shorter than for SPME [21]. The LPME extraction time is relatively long, but with the magnetic stirring table used, up to 10 replicates could be extracted simultaneously overnight and analyzed by HPLC the following day.

Parallel to the extraction of LAS:THA ion pairs, there is also a continuous transport of H⁺ to the acceptor, since the amine is transported in charged form. Eventually, this decreases the acceptor pH and reduces the driving force for the extraction. Previously, 0.01 M NaOH was used as acceptor in SLM extractions, which lasted ∼40–50 min [24]. However, using HFs and employing longer extraction times, E_e dropped after 1–5 h. Therefore, 32.5 mM phosphate buffer adjusted to pH 12.0 was used, which gave more stable pH and E_e. By doubling the buffer concentration to 65 mM, higher E_e were obtained, which gave the extraction equilibrium in Figure 1. 130 mM phosphate buffer was also tested (n = 3 at 15 h), but this lead to that E_e dropped with ∼85% for all homologues. Possibly, a buffer with lower ionic strength and density could have improved extraction more than phosphate, due to lower salting out-effect and faster diffusion. Here, 65 mM phosphate buffer was considered optimal.

As is noted in Figure 1 the extraction is slow. Typically, equilibrium times in LPME are faster and range from ∼30 min [23,33] to 6–7 h [25]. A slow transfer of LAS over the membrane/acceptor interface was previously suspected in SLM extraction of LAS [24]. The long extraction time observed here is also similar to the slow extraction in ion-pair LPME of a cationic amine surfactant [34]. The extraction of cationic surfactant was performed in a 2-phase system, which lacked the SLM/acceptor interface. This suggests that the rate-limiting step in ion pair HF-LPME of surfactant is perhaps either in the transfer of analyte across the sample/SLM phase boundary or into the bulk of the SLM, and possibly not at the interface between the SLM and the acceptor.

Slow extraction could possibly be due to extraction of reversed micelles. The positive counter ion could balance the repulsive effect between sulfonate head groups. Reversed micelle extraction of proteins requires extraction times of 24 h [35] to 100 h [36]. In surfactant enhanced LPME of drugs, maximum enrichment was reached after ∼40 min, but it was found that when the sample concentration of (nonionic) surfactant exceeded the critical micelle concentration, E decreased sharply [33]. The interpretation was that drugs were incorporated into the micelles, which could not completely pass the HF pores. However, data on reversed micelles is more scarce than of micelles in aqueous solution [3]. Micelle formation in water usually occurs over a limited concentration range, while physical properties of non-aqueous solutions related to (reversed) micelle formation often undergo a continuous transition over orders of magnitude in concentration [3].

More vigorous stirring of the sample increases the contact between the sample and the HF, decreases the boundary layer on the HF outside and thus increases enrichment. 330, 770 and 1100 rpm were tested in 100, 300 and 500 mL samples extracted in E-flasks (n = 1 for each speed:volume combination). For 1100 rpm, E_e was very low or the extract was washed out, which was noticed when it should be collected with a syringe. The highest E_e was observed for 300 mL stirred at 770 rpm, followed by 500 mL at 770 rpm. In volumetric flasks, 250 and 500 mL gave about the same E_e (Figure 2), but the precision was much better for 250 mL (RSD of 2–13% for the different homologues) than for 500 mL (RSD 30–67%). In retrospect, too few replicates may have been performed for 500 mL. If one of the extracts for 500 mL was a statistical outlier, this would have decreased precision. This experiment was performed with the setup using metal rods and not syringes holding the HFs during the extraction. 250 mL was considered optimal at that time and thereafter used throughout.

It was observed that precision depended on how much the HF moved as the sample was stirred. In 100 mL volumetric flasks, 550 rpm was the lowest stirring speed that did not cause the HF to shake in an unstable way. Table 1 compares extraction between 250 mL × 770 rpm and 100 mL × 550 rpm. Typically, increasing the sample volume in membrane extraction increases the enrichment, E_e, but decreases the fraction of extracted analyte from the sample, E. This could also be observed here. However, for the lower homologues C₁₀ and C₁₁, E was higher for the larger sample. It can also be noted that the difference in both E_e and E was lower for all homologues for 250 mL than for 100 mL. When the fiber is exposed to less analyte, higher homologues are favored. Whether this is at the expense of the lower homologues is unclear.

Regardless of sample volume, it was observed that enrichment was not constant over the studied sample concentration range. In the interval 5–1000 μg L⁻¹, linear R²-coefficients for the acceptor concentration ranged from 0.934 (C₁₀) to 0.9745 (C₁₃) and the intercepts were significantly below zero. Better correlations were obtained by fitting the data to 3^rd order equations, giving R²-coefficients from 0.9925 (C₁₀) to 0.9969 (C₁₂) (Figure 3).

It was found that for lower sample concentrations (<150 μg L⁻¹), the degree of enrichment was reversed compared to higher sample concentration (1 mg L⁻¹), i.e., for low concentrations C₁₀ was enriched the most and C₁₃ the least. As the sample concentration increased, the higher homologues were eventually enriched more. C₁₁ surpassed C₁₀ about 150–200 μg L⁻¹, C₁₂ surpassed C₁₁ ∼250 μg L⁻¹ and C₁₃ surpassed C₁₂ about 350–400 μg L⁻¹ (100 mL samples, 20 h at 550 rpm). Furthermore, here it was also observed that the equilibration time was higher for 100 mL samples than for 250 mL samples. Unlike the equilibrium observed after 15 h for 250 mL samples, the enrichment was increasing for all homologues, except possibly C₁₀, even at 24 h.

Non-linear extraction in ion-pair LPME of (cationic) surfactants (between 1 and 60 μg L⁻¹) was also recently observed [34]. This was explained by adsorption of the cationic surfactants on the glass. However, anionic surfactants such as LAS are not expected to adsorb on glass or silica at pH 7, due to the negative charge of both silica surface and LAS [37]. This leads to the question whether the enrichment in ion pair LPME of surfactants is dependent on analyte concentration. In reversed micelle assisted protein extraction, it was reported that at low surfactant concentrations, mass transfer over the membrane interface limited extraction, while at higher surfactant concentrations, mass transfer through the membrane was limiting [37]. In extraction of polyamines, it was observed that E_e increased when the surface tension toward the acceptor decreased (by increasing the carrier concentration or decreasing the acceptor pH) [25]. Possibly, THA decreases the surface tension likewise here. In the current work, after the LAS:THA ion pair has been broken up in the acceptor, neutral THA may be back-extracted and diffuse back to the sample side of the membrane, where THA can once again be protonated. At any given time, THA back-extraction would depend on the already achieved E_e. LAS ion pair formation and extraction with THA molecules that are already situated at the sample/membrane interface is probably faster than likewise pair formation and extraction from the bulk of the sample solution. Thus, it overall seems reasonable that enrichment in ion-pair LPME of LAS is dependent on sample concentration.

A non-linear curve may appear linear in a limited range, i.e., the studied method might find practical use in a more limited concentration range. In the lower interval 10–50 μg L⁻¹, which in fact is more environmentally relevant, better linear correlations were obtained (Figure 4). It may be noted that the R²-coefficients decreased with increasing homologue number; 0.9940 for C₁₀, 0.9933 for C₁₁, 0.9561 for C₁₂ and 0.0811 for C₁₃. E_e was 152 for C₁₀, 135 for C₁₁ and 64 for C₁₂. For C_13, no reliable E_e could be determined. C₁₃ also elutes last in the HPLC, exhibiting the largest band broadening. Average E in this interval was 0.63% (C₁₀), 0.56% (C₁₁) and 0.27% (C₁₂) and 0.20% (C₁₃) (which is rather low).

When the ionic strength of water is increased, the partitioning of hydrophobic solutes into lipophilic phases increases [2], but attraction between any pair of specific ions decreases. On adding up to 10% (w/v) NaCl to the sample, E_e of C₁₀ increased slightly while the extraction of C₁₂ and C₁₃ decreased (Figure 5). C₁₁ was approximately constant. Further extractions were made without addition of NaCl.

In tap water spiked to 1 mg L⁻¹, all homologues were negatively affected (Table 2). However, C₁₀ was affected the most and C₁₃ the least. This was the opposite trend to the effect of NaCl. Anionic surfactants precipitate partly with divalent cations [2] and the local tap water contained 21–22 mg L⁻¹ calcium and 1–2 mg L⁻¹ magnesium (as measured by flame atomic absorption), which could influence the extraction. However, LAS sensitivity to water hardness usually increases with homologue number [2].

LAS can bind to dissolved organic matter (DOM), which decreases the freely dissolved concentration and affects the bioavailable amount [38,39]. It was observed that lower concentrations could be extracted when surface water samples were filtered prior to spiking (data not shown). The turbidity of the environmental samples was visually similar to that of 2.5 mg L⁻¹ solution of Aldrich humic acids. Turbidity also depends on inorganic colloids, but in a simplified approach, the effect of organic matter on extraction was tested at 2.5 and 5 mg L⁻¹ of humic acids. Based on reported log K_DOC for LAS to Aldrich humic acids_, the bound fraction (for isomers substituted at the second alkyl carbon) at 2.5 mg L⁻¹ of humic acids would be about 4% for C₁₀, 7% for C₁₁, 17% for C₁₂ and 44% for C₁₃ [39]. Here, a larger effect was observed at LAS concentrations <100 μg L⁻¹, where the average relative E_e compared to humic acid free sample was ∼60% for C₁₀–C₁₂ and 34% for C₁₃ (Table 2). Between 100 and 500 μg L⁻¹ LAS there was a slight decrease in enrichment due to humic acids, but no significant effect was seen above 750 μg L⁻¹ (Figure 6).

Quantification of LAS in surface water was attempted by standard addition from 5–50 μg L⁻¹. In spiked environmental samples, E_e was about 7–19% of E_e in buffered reagent water (comparing Figure 4 and Table 2). The decrease may be due to matrix effects such as DOM (Table 2) or divalent cations, as the river water contains about 75–85 mg L⁻¹ calcium and 7–8 mg L⁻¹ magnesium (previously determined by titration). Precision was evaluated by extraction of five STP effluent samples spiked to 10 μg L⁻¹. For C₁₀, C₁₁ and C₁₂, precision was quite similar to buffered reagent water (Tables 1 and 2). Method LOD for C₁₀, C₁₁ and C₁₂ were estimated to about 5 μg L⁻¹, for each individual group of homologues and based on nominal LAS concentration. Taking into account the estimated fractions of homologues, this gives individual homologue LODs ∼1 μg L⁻¹. C₁₃ had higher method LOD, due to its generally inefficient extraction at concentrations <50 μg L⁻¹ and band broadening. C₁₃ could be barely detected as very small peaks in an effluent sample nominally spiked to 25 μg L⁻¹. This extract chromatogram is presented in Figure 7, together with a chromatogram of a standard to illustrate the general separation.

For the effluent, C₁₀ and C₁₁ were detected in all non-spiked aliquots and C₁₂ in five of six such replicates. For the river samples, only one of the non-spiked samples (after the STP) contained LAS. However, all these concentrations were too close to LOD for certain quantification.